TL;DR
This paper introduces a novel framework for pricing advertising options by accounting for discontinuous spot market prices and using an average price over a future period for option payoffs. This approach offers a more realistic and generalized model for online advertising contracts, moving beyond previous limitations of continuous price assumptions and single-point valuations.
What Problem Does This Paper Address?
Online advertising platforms often sell ad inventory through real-time auctions. While these auctions are dynamic, they can lead to unpredictable costs for advertisers and volatile revenue for publishers. Advertising options offer a middle ground: a contract giving buyers the right, but not the obligation, to purchase ad inventory at a pre-specified price in the future. However, existing models for pricing these options have two major shortcomings: they typically assume the underlying spot market price evolves continuously (which is not true in reality, as prices can jump or spike) and they calculate the option's payoff based on the price at a single future point in time (leading to biased valuations). This paper aims to address these limitations by developing a more robust pricing model.
Methods
The authors propose a new advertising option pricing framework with two key innovations:
- Path-Dependent Payoff: Instead of relying on a single price point, the option's payoff is calculated based on the average spot market price over a specific future period. This average is measured using the "power mean," a flexible metric that encompasses common averages like arithmetic, geometric, and harmonic means, as well as minimum and maximum values. This makes the option "path-dependent," meaning its value depends on the entire price history within the averaging period.
- Jump-Diffusion Stochastic Models: To capture the discontinuous nature of online advertising spot prices, the paper employs jump-diffusion stochastic processes. These models combine continuous price movements (like Brownian motion) with sudden jumps and spikes, better reflecting real-world market behavior.
A general pricing algorithm is developed using Monte Carlo simulation. Additionally, an explicit pricing formula is derived for a special case where the payoff is based on the geometric mean.
Datasets
The study utilizes three real-world datasets:
- SSP Dataset: A display advertising dataset from a medium-sized supply-side platform (SSP) in the UK.
- Google UK Dataset: A sponsored search dataset for keyword auctions.
- Google US Dataset: Another sponsored search dataset for keyword auctions.
These datasets were used for training, testing, and investigating the statistical properties of spot market prices in online advertising auctions. The paper also provides a comprehensive summary of "stylized facts" observed in these datasets, which are common statistical properties of market prices.
Key Insights and Ideas
- Realism in Pricing: By incorporating discontinuous price movements and average-based payoffs, the model offers a more accurate reflection of how online advertising markets actually function.
- Flexibility of Power Mean: The power mean is a powerful tool that unifies several existing option payoff structures, making the proposed framework a generalized approach.
- Stylized Facts of Online Advertising: The paper meticulously documents empirical properties of spot market prices, such as jumps, spikes, non-normality, heavy tails, and absence of autocorrelations, providing valuable insights for future modeling efforts.
- Impact of Market Conditions: The study highlights how market performance (bull vs. bear markets) and pricing specifications (ITM, ATM, OTM) significantly influence seller revenue when offering advertising options.
Key Results
- The proposed jump-diffusion models effectively capture stylized facts like jumps, spikes, non-normality, and heavy tails observed in online advertising spot prices.
- The Monte Carlo simulation and derived explicit formula provide a robust method for pricing these average-price advertising options.
- In bull markets, selling advertising options can lead to lower seller revenue compared to auctions, as buyers are more likely to exercise options to hedge against rising prices.
- In bear markets, advertising options can significantly increase seller revenue, as buyers are less likely to exercise, and the seller benefits from the upfront option premium.
Limitations and Future Work
The authors acknowledge several limitations and suggest future research directions:
- Constant Volatility: The current model assumes a constant volatility term. Future work could explore stochastic-volatility jump-diffusion models to account for time-varying volatility.
- Cyclical Patterns: While the jump-diffusion model can reproduce some cyclical patterns, more accurate models could be developed to explicitly incorporate them.
- Capacity Constraints: The current framework assumes unlimited inventory. Future research could incorporate capacity constraints and analyze the strategic implications for both buyers and sellers.
- Volatility Clustering: The model does not explicitly account for volatility clustering. Incorporating this property could further enhance the model's realism.
Summary updated